<html><head><meta http-equiv="Content-Type" content="text/html charset=us-ascii"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><br><div><div>On Apr 16, 2012, at 3:30 PM, Chandler Carruth <<a href="mailto:chandlerc@google.com">chandlerc@google.com</a>> wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div style="font-family: Optima; font-size: medium; font-style: normal; font-variant: normal; font-weight: normal; letter-spacing: normal; line-height: normal; orphans: 2; text-align: -webkit-auto; text-indent: 0px; text-transform: none; white-space: normal; widows: 2; word-spacing: 0px; -webkit-text-size-adjust: auto; -webkit-text-stroke-width: 0px; ">Does sinking these into the DAGCombine layer help? How much does it break?</div></blockquote></div><br><div>I tried disabling just the InstCombine transforms that hide shl instructions behind bitmasks. Even though DAGCombine has the same transforms, it causes some pretty bad regressions:</div><div><br></div><div><div><font face="Menlo"><span style="font-size: 13px;">External/SPEC/CINT95/147_vortex/147_vortex                                   0.294   0.322   +9.6%  +40mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">MultiSource/Benchmarks/Olden/tsp/tsp                                         0.680   0.748   +9.9%  +41mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">SingleSource/Benchmarks/Shootout-C++/except                                  0.116   0.128  +10.8%  +45mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">SingleSource/Benchmarks/Shootout/strcat                                      0.102   0.113  +11.1%  +46mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">SingleSource/Benchmarks/Shootout-C++/hash                                    0.455   0.507  +11.4%  +47mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">External/Povray/povray                                                       2.015   2.246  +11.5%  +47mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">External/SPEC/CINT2000/255_vortex/255_vortex                                 1.814   2.044  +12.7%  +52mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">SingleSource/Benchmarks/Shootout-C++/heapsort                                1.871   2.132  +13.9%  +57mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">SingleSource/Benchmarks/Shootout-C++/ary3                                    1.087   1.264  +16.3%  +65mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">MultiSource/Benchmarks/SciMark2-C/scimark2                                  27.491  23.596  -14.2%  -66mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">MultiSource/Benchmarks/Olden/bisort/bisort                                   0.360   0.428  +19.0%  +75mB</span></font></div><div><font face="Menlo"><span style="font-size: 13px;">MultiSource/Benchmarks/Olden/bh/bh                                           1.074   1.287  +19.9%  +79mB</span></font></div></div><div><br></div><div>(Running on Sandy Bridge, x86-64)</div><div><br></div><div>I'll try to figure out why.</div><div><br></div><div>/jakob</div><div><br></div></body></html>